Date:   Wed, 24 Aug 2016 13:06:45 +0800
From:   Baoquan He <bhe@...hat.com>
To:     Jonathan Corbet <corbet@....net>,
        "Zhou, Wenjian/周文剑" 
        <zhouwj-fnst@...fujitsu.com>
Cc:     linux-kernel@...r.kernel.org, akpm@...ux-foundation.org,
        dyoung@...hat.com, vgoyal@...hat.com, kexec@...ts.infradead.org,
        linux-doc@...r.kernel.org, xlpang@...hat.com, joe@...ches.com
Subject: Re: [PATCH v9 1/2] Documentation: kdump: remind user of nr_cpus

On 08/22/16 at 09:14am, "Zhou, Wenjian/周文剑" wrote:
> On 08/19/2016 11:57 PM, Jonathan Corbet wrote:
> >On Fri, 19 Aug 2016 08:33:21 +0800
> >"Zhou, Wenjian/周文剑" <zhouwj-fnst@...fujitsu.com> wrote:
> >
> >>I was also confused by maxcpus and nr_cpus before writing this patch.
> >>I think it is a good choice to describe it in kernel-parameters.txt.
> >>
> >>Then, only two things need to be done I think.
> >>One is move the above description to maxcpus= in kernel-parameters.txt.
> >>And the other is replace maxcpus with maxcpus/nr_cpus in kdump.txt.
> >>
> >>How do you think?
> >
> >That is not quite what I had in mind, sorry.  What I would really like to
> >see in kernel-parameters.txt is an explanation of how those two parameters
> >differ - what do they do differently and how should a user choose one over
> >the other?  What we have now offers no guidance in that matter.
> >
> 
> I thought about it. I think user may not need this.
> What user really want to know is how to choose.
> And it is also not a hard work. If nr_cpus is not supported by the ARCH, use maxcpus.
> Otherwise, nr_cpus. The reason why maxcpus still exists is nr_cpus can't be supported
> by some ARCHes.

I think Jon is suggesting that a note can be added into
kernel-parameter.txt to tell what's the difference between nr_cpus and
max_cpus. I checked code and discussed within our kdump team, max_cpus
is used to limit how many 'present' cpus are allowed to be brought up
during system bootup, while nr_cpus is used to set the upper limit of
'possible' cpus. E.g on my laptop, there are 4 cpus while 4 hotplug
cpus, altogether 8 possible cpus. Possible cpus slot is for cpu hot
plug, means during bootup you want to bring up 4 present cpus, but
later you could physically hot plug 4 others. Because of attribute of
some static percpu variables, we need pre-allocate memory for all
possible cpus though some of them may not be really used if no extra
cpu physically hot plugged after system bootup.

Hence for kdump kernel, people never want to do a cpu hot plug in there.
That's why we want to use nr_cpus to limit the number of possible cpu to
save memory. E.g still on my laptop, if I want to do a kdump, the number
of possible cpu is still 8, but you may want to use only 1 cpu to dump,
maybe 2 or 3 for parallel dumping. But you absolutely don't want to set
nr_cpus=8 in your kdump kernel cmdline, though it doesn't cause failure,
memory is wasted because of percpu pre-allocation. So specifying nr_cpus=1
is much better. While with specifying max_cpus=1, the number of possible
cpu is still 8. That's the reason. On x86_64 and s390, there's another
kernel para "possible_cpus=xx" which can be used to set possible cpus for
cpu hot plug. Only when "possible_cpus=0" is specified, smp is disabled.
I am not very sure why this is introduced, number of possible cpu is
decided by the min value of nr_cpus= and possible_cpus=.

nr_cpus and maxcpus might not be very clear to people which are
described in Documentation/kernel-parameters.txt.

Hi Jon, do you think change as below is OK to you?


>From 8b940193a29acf0857d4975d77f4b9f48e2d6cb8 Mon Sep 17 00:00:00 2001
From: Baoquan He <bhe@...hat.com>
Date: Wed, 24 Aug 2016 11:14:34 +0800
Subject: [PATCH] docs: kernel-parameter : Improve the description of nr_cpus
 and maxcpus

>From the old description people still can't get what's the exact
difference between nr_cpus and maxcpus. Especially in kdump kernel
nr_cpus is always suggested if it's implemented in the ARCH. The
reason is nr_cpus is used to limit the max number of possible cpu
in system, the sum of already plugged cpus and hot plug cpus can't
exceed its value. However maxcpus is used to limit how many cpus
are allowed to be brought up during bootup.

Signed-off-by: Baoquan He <bhe@...hat.com>
---
 Documentation/kernel-parameters.txt | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)

diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index 46c030a..25d3b36 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -2161,10 +2161,13 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
 			than or equal to this physical address is ignored.
 
 	maxcpus=	[SMP] Maximum number of processors that	an SMP kernel
-			should make use of.  maxcpus=n : n >= 0 limits the
-			kernel to using 'n' processors.  n=0 is a special case,
-			it is equivalent to "nosmp", which also disables
-			the IO APIC.
+			will bring up during bootup.  maxcpus=n : n >= 0 limits
+			the kernel to bring up 'n' processors. Surely after
+			bootup you can bring up the other plugged cpu by executing
+			"echo 1 > /sys/devices/system/cpu/cpuX/online". So maxcpus
+			only takes effect during system bootup.
+			While n=0 is a special case, it is equivalent to "nosmp",
+			which also disables the IO APIC.
 
 	max_loop=	[LOOP] The number of loop block devices that get
 	(loop.max_loop)	unconditionally pre-created at init time. The default
@@ -2773,9 +2776,12 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
 
 	nr_cpus=	[SMP] Maximum number of processors that	an SMP kernel
 			could support.  nr_cpus=n : n >= 1 limits the kernel to
-			supporting 'n' processors. Later in runtime you can not
-			use hotplug cpu feature to put more cpu back to online.
-			just like you compile the kernel NR_CPUS=n
+			support 'n' processors. It could be larger than the
+			number of already plugged CPU during bootup, later in
+			runtime you can physically add extra cpu until it reaches
+			n. So during boot up some boot time memory for per-cpu
+			variables need be pre-allocated for later physical cpu
+			hot plugging.
 
 	nr_uarts=	[SERIAL] maximum number of UARTs to be registered.
 
-- 
2.5.5

Powered by blists - more mailing lists